6/13/2025 - Mastering ChatGPT Prompts: A Deep Dive into Project Workflow

Description

Chapters

Transcript

0:01 What do you think, boy? What do you
0:03 think?
0:06 [Music]
0:23 You and I all alone.
0:28 Sunday morning here at home.
0:35 [Music]
0:41 You and I here alone
0:46 Sunday morning here at home.
0:51 Blue and the coffee strong. It's true.
0:56 [Music]
0:58 Then I open my eyes to a dream realized
1:02 in front of me.
1:06 And I haven't got a clue what in the
1:09 world is happening to me.
1:14 Think I think I'm happy
1:17 like first day summer vacation. Happy.
1:21 Got to get a little rest and relaxation.
1:24 Happy
1:25 like the choir on Sunday morning singing
1:28 true.
1:31 [Music]
1:44 Friday night date night at the AI
1:46 learning lab. Are we going to learn
1:48 tonight? No. We going to have a date
1:51 tonight? Yes,
1:53 we're going to have fun.
1:56 You never know. You never know. At the
1:59 learning lab, sometimes there's
2:01 learning. Sometimes there's comedy and
2:03 entertainment. A lot of times there's a
2:05 dog.
2:07 And then we've got Champy, too.
2:10 Get it? I was calling myself a dog.
2:17 Um,
2:21 that was comedy explained.
2:23 See, we're we're when we're when we're
2:25 teaching here, we're not just teaching
2:27 AI. We're teaching comedic structure.
2:31 We're pointing out the obvious. We're
2:34 making dad jokes about dad jokes.
2:43 All right. Fantastic.
2:48 If it weren't for for Friday night date
2:50 night, I wouldn't have a date at all.
2:52 Oh, Kelly, you're working too hard. You
2:55 got too much AI in your life. We got to
2:57 get you got to get you some of the
2:59 people. Hey, Sharon Crawford, what's
3:01 happening?
3:03 Oh, Kevin Stewart, does he always make
3:06 jokes like this? Exactly. Yeah. Hey, you
3:09 know about this channel that you're
3:10 watching? I No, I went No, I did. I I
3:13 watched it last night. I did was this it
3:17 was the AI learning lab with the
3:21 rotund the rounder gentleman the older
3:24 round gentleman. Yeah. Yeah. No, I did.
3:27 I saw it. It was um
3:30 Yeah, it was um it was interesting. It
3:34 was interesting. Yeah. Mhm. And you
3:36 watch that every night? Yeah. Yeah. No,
3:39 I think No, I think for you that's
3:41 great. I think it's great for you. Yeah.
3:43 No, it's it's it's I just you know what
3:46 it is. I probably don't like AI. That's
3:48 probably what it is. I just
3:52 Good evening everybody. All right.
3:58 [Music]
4:15 [Music]
4:24 Champy went out on a walk today. It was
4:26 very traumatizing apparently because
4:28 Champy got in a fight the other day
4:30 because he's a bully, but he got his ass
4:32 kicked by by a uh not a go. What was it?
4:37 A golden retriever? What was it that
4:38 kicked your ass? It was some dog that
4:40 looked pretty, but he he looked prettier
4:43 when he had his jaws around your neck.
4:46 Right, you big dumb dumb. Why'd you go
4:50 pick on that dog? Huh? Now he's all
4:53 scared today.
4:57 Dogs are funny.
5:01 Oh, Champy, what are we going to do with
5:03 you?
5:04 Oh, Vicki. And no dirty nachos for me
5:07 today.
5:10 Oh, you had Sichuan and snooze. That's
5:13 funny.
5:17 I love that for you. It was one of my
5:18 least favorite phrases. Oh, I love that
5:21 for you. Oh, you like that? Yeah. No, I
5:22 love that for you. That's fantastic.
5:25 Yeah. No, it's for you. For you that's
5:27 awesome. That's awesome. It's just, you
5:29 know, I just I have a different
5:31 aesthetic. Maybe it's just you know I
5:34 when I think of education I think of
5:36 more professorial kind of experiences if
5:41 you with a curriculum and such you know
5:43 with you know actual learnings. Yeah
5:46 that because the AI learning lab was um
5:50 not that I mean it was it was an
5:52 alternative approach to education which
5:54 I think is innovative. It was super
5:57 innovative.
5:59 Super innovative. Yeah. No, it was No,
6:01 it was really I think you have a life.
6:04 Yeah. And it's And you're choosing to do
6:06 things in it and I just for you I think
6:09 that's really special. Yeah. No, I No,
6:11 it's it's all good. It's all good. I
6:13 wasn't offended. There was a lot of
6:15 cussing there. There was there was a lot
6:17 of cussing. Yeah. No, but it I But you
6:20 know, I've got thick skin. I've I've
6:22 certainly got thick skin for for you
6:24 know, people who choose to be that way.
6:26 No, I I you know, they choose to live
6:29 their life and and everyone should have
6:31 a space that you know I don't
6:34 necessarily need to live in.
6:36 You know, you know how that is, right?
6:38 Yeah. Yeah. No, no, I do I do think
6:41 we're good friends. Yep. No, I do. I
6:43 Yeah, I I I I know more about you now
6:46 than I did than I did yesterday. So,
6:48 that's There's always a positive.
6:50 There's always a positive. All right.
6:55 I don't know why this turned into a an
6:58 improv of of one of the characters from
7:01 Office Space, but you know,
7:06 thank you for sharing the live Danielle
7:08 on a Tik Tok and whoever else has done
7:10 that. I see some other people have done
7:11 that. Thank you very much, Silver Fox,
7:14 as always. Kyle, what's up? What's
7:16 happening? Kie,
7:20 what is shaking?
7:24 When someone asks, "Does he have any
7:25 qualifications?" Yeah, about that.
7:28 [Music]
7:53 Girl, you're looking
7:55 [Music]
7:57 find a night.
8:01 Every fellas got you in his sight.
8:08 What you doing with a clown like me?
8:14 Wait.
8:16 Surely
8:19 one of life's little
8:22 mysteries.
8:24 So tonight
8:27 [Music]
8:29 stars above
8:33 how did I ever win your love?
8:41 Uh,
8:43 wait.
8:46 I don't remember the chords. That's a
8:48 pretty song that I can't quite sing.
8:49 It's too high for me. It's very high.
8:52 It's very high. Very, very high. Um, so
8:58 a bless him. Exactly. That's exactly
9:01 what it is, Vicki. It's good for Isn't
9:04 that sweet for you? It's totally bless
9:05 him. A bless him. Ain't that ain't that
9:08 sweet? you know, that AI learning lab,
9:09 you know, bless him. Bless him. He's out
9:12 there night after night, um, you know,
9:15 attempting to express himself. Just
9:17 bless him, right? You know, to each
9:20 their own. To each their very own. Oh,
9:23 yes. No, it was it was an experience. It
9:27 was absolutely an experience. Yes. No.
9:30 No. It's Yeah. I Yeah. No, it was it was
9:34 a Have I tried chat GPT? No. No, no, not
9:38 yet. Not yet. Not yet. Not yet. I You
9:41 know,
9:44 and you know, we just got rid of the fax
9:45 machine, so slowly progressing. You
9:49 know, you know what I Yeah. No, still
9:51 have the landline. Yeah, we still have a
9:53 real phone. Okay. Yep. No, we still
9:56 have, you know, because I'll tell you
9:57 what, when the power goes out, we can
9:59 still make phone calls. Okay. All right.
10:02 That's why we have that.
10:05 Oh, yes. Yes. Yes. Yes. Yesas.
10:09 Now we know how to invoke the other
10:11 Kyle.
10:15 [Music]
10:18 [Applause]
10:19 [Music]
10:23 Um, happy Friday night everyone. So, uh,
10:27 so I I kind of have a plan tonight,
10:30 which don't get me wrong, it's my
10:32 version of a plan, so there's no actual
10:34 plan. There's no actual curriculum, but
10:38 so two nights ago, maybe two or three
10:41 nights ago, we made the Bigfoot video.
10:44 Someone came in and asked about how do
10:45 you make the Bigfoot video? And then I
10:48 said, you know, I started writing a
10:51 prompt using my big fat brain. And then
10:53 I was like, oh, wait a minute. Don't use
10:55 your brain. Have Chat GPT help you write
10:57 that. And it did. And it came up with
10:58 something really good. So I thought it
11:01 would be cool today. So on office hours
11:04 today and then on the AI salon
11:07 mastermind call which is right after it
11:10 um
11:12 we had a lot of people talking and
11:14 sharing really kind of deep
11:17 deep uses of how they're doing
11:19 prompting. Um and like Sam is it Sam?
11:23 What's Sam's last name? Sam it's not Sam
11:25 Elliott is it? Is it Sam Elliott? That's
11:27 an actor. But do we also have a Sam
11:29 Elliott? What's Sam's last name?
11:30 Whatever. It doesn't matter. Sam was
11:32 talking about this 16page prompt that
11:34 he's working on for work and I was like
11:36 I like the concept of a one-page prompt
11:40 intimidates me and he's got this 16page
11:42 prompt. So he was giving us all these
11:43 best practices about how you do that.
11:47 And then I think it was Vicki or someone
11:49 else was talking about um doing prompt
11:53 generation within a project because you
11:55 can keep adding documents and chats to
11:58 the project that contribute to it. So, I
12:00 thought what might be cool
12:04 is to go into chat GPT,
12:08 have it come up with prompt frameworks,
12:12 put those in a canvas, export those as
12:15 PDFs. So basically have it give itself
12:18 instructions for how to write prompts
12:21 for things like scenes, characters,
12:24 voice designs, things that we can, you
12:27 know, start to stitch together
12:28 components to make videos
12:31 and then put that all in a project and
12:33 then we'll have a place where we can go
12:35 start writing prompts and then we can
12:37 pull up things like 11 Labs voice
12:40 designer or we can um pull up like Hedra
12:43 or Runways character animation things
12:45 and like runway you can combine a
12:49 location, some characters and
12:54 oh and like a a a storyboard like a
12:57 storyboard layout. So I just thought it
12:58 would be cool to make some sort of
13:00 component maker. So we can try that.
13:03 Now, what I know is that whenever I go
13:06 in with some sort of intention like
13:08 we're going to do something that's a
13:09 that's a real thing, it gets really
13:11 boring really fast cuz AI still sucks.
13:19 But we'll try that. Okay.
13:22 It is indeed Sam Samuel Elliott.
13:25 Beautiful.
13:27 He is a a a very intelligent, driven,
13:30 passionate young man. I like Sam a lot.
13:33 [Music]
13:40 Um Gareth, who's in a regular who comes
13:42 in here a lot, he said today that he's
13:45 um he's gotten a new gig where he's like
13:50 an AI he he basically
13:53 takes has taken his product development
13:57 background and because he knows AI, he
14:00 was given this sort of innov innovative
14:03 product development role that's about
14:05 using AI to improve this company's
14:07 products. So, further evidence of get
14:11 your ass curious about AI and and sort
14:14 of bolted on to what you your core
14:16 expertise and I think if you put that
14:19 combination in the marketplace, it's
14:21 you're a different kind of hire right
14:23 now, which is really cool.
14:26 Kevin Stewart, I fly out to Japan to
14:29 Japan for a conference for AI next week.
14:31 Oh, that's cool. That's awesome. That's
14:33 really cool.
14:35 All right, Francia.
14:37 Have you found chat GPT's answers over
14:40 time are reflecting your own own inputs
14:42 and biases? Yes, Kie. So,
14:46 if you've noticed that in the past six
14:48 weeks or so, that's because about six
14:50 weeks ago, maybe two months ago, Open
14:52 AAI added um memory.
14:57 Um,
14:58 so chat GPT now can remember I think
15:02 it's your last year of chats. It knows
15:05 them all.
15:07 And if you're chatting, it will often
15:11 invade your current chat with stuff
15:13 you've talked about before, which for a
15:15 channel, for an account like mine is
15:17 really hard because I show so many
15:19 different things on this channel that,
15:21 you know, my my chat is very polluted
15:24 with with idiocy on my part. Um,
15:31 so if you want to get rid of that, you
15:33 can do it. Here, let me show you how to
15:34 do that. Hang on a sec. Let me get get
15:38 things right here.
15:40 I am This is the last night without
15:42 producer Brandon and I think I have to
15:43 turn on the air conditioner because it's
15:44 getting hot in here and I might pass
15:46 out. Okay, so here's Chat GPT.
15:51 All right. In the upper right hand
15:53 corner,
15:54 go to your account and then go to
15:57 settings
15:58 and then in here, choose personalization
16:01 and you can turn off memory.
16:05 There's two there's two kinds of memory.
16:08 There's saved memories, which this was
16:10 the old memory. The old memory was you
16:13 could explicitly say, "Hey, chat GPT
16:15 remember this thing." And it would write
16:16 it to this file. And then sometimes it
16:19 would just it it would kind of decide to
16:21 to save a memory for you. So sometimes
16:23 you would write a prompt and it would
16:25 say memory saved. You can go in and you
16:28 can edit that one. This one below it,
16:31 reference chat history. This is the new
16:33 one. You can turn that off discreetly.
16:36 Um so if you want chat GPT to not be
16:40 influenced and biased by your previous
16:42 chats, that's how you turn that stuff
16:43 off. All right.
16:46 So there you go. Look, the actual
16:47 education here, people, you see what I'm
16:50 saying? It's not just a waste of time,
16:51 this channel.
16:54 It's mostly a waste of time. But it's
16:56 not all It's not all a waste of time.
16:58 Just mostly.
17:03 [Laughter]
17:07 Oh, lordy, lordy, lordy, lordy, lordy.
17:11 Do you know any songs? I Yeah, I know. I
17:15 know. I know the the uh the 12 songs. I
17:19 know the 12 songs of Christmas. Um, I
17:22 know lots of songs in my iPad, but the
17:25 ones I actually remember are
17:29 [Music]
17:32 [Applause]
17:34 [Music]
17:38 this place I can rest my paid
17:45 to gather my thoughts in sweet silence.
17:52 And this is this place where the
17:54 feelings on dead
17:56 [Music]
17:58 from an overexposure to violence.
18:03 This is a place I can slowly face. The
18:06 only one I truly can know.
18:12 These are tears from a long time ago.
18:15 Got these tears from a long time ago.
18:19 I need to cry. That is a song. These are
18:24 tears for a long time.
18:29 Go.
18:35 Um,
18:36 caveat. I only know depress depressing
18:40 songs.
18:41 That one's called 30 Years of Tears by
18:44 John Hyatt. incredibly prolific,
18:46 talented songwriter, performer, amazing
18:50 amazing talent. Uh most of his songs are
18:54 just they'll just drive you to drink.
19:01 Oh man, when do you start taking
19:03 requests? When I learn more than 12
19:05 songs.
19:08 What education? I'm out. I know, Kelly.
19:10 That's not why we're here, is it?
19:13 Okay. Um,
19:19 the other thing I thought I I had a very
19:21 busy day today. Busy, busy, busy. I was
19:23 a busy boy. So, normally I try to find
19:27 an hour or two during the day to surf
19:29 Twitter and see if anything launched
19:31 that is is worth looking at. And I
19:34 didn't do that. So, it might be fun to
19:35 just just go scroll Twitter for a bit
19:38 and see if there's anything interesting
19:39 there.
19:41 Um,
19:42 M I'm here strictly for the
19:45 entertainment which that that's saying a
19:47 lot on on many levels.
19:50 I took a pill in a beza. Nice. Joy
19:54 party's in the house. Joy party in the
19:57 house.
19:59 We could make images. We can make
20:01 images. Well, again like what I want to
20:04 do I I am my I am newly
20:08 I'm newly committed
20:12 [Music]
20:17 I'm starting to get my head around. I've
20:19 been saying this for a while. 2025 is
20:21 the year I'm going to stop using my
20:23 brain.
20:24 [Music]
20:26 [Laughter]
20:29 And I want to put some context around
20:31 that.
20:35 [Music]
20:44 So, okay. So, let me give you my theory
20:46 on on on
20:49 creative energy.
20:52 Let me turn on the air conditioner
20:53 first. Hang on.
20:59 All right.
21:01 So, one of the things I've learned over
21:03 my many, many years, and this is true
21:06 for me, this might not be true for
21:08 everyone. I've got ADHD. I'm neurospicy.
21:11 So, that might have something to do with
21:12 it. But
21:14 my experience with creativity
21:18 is that on any given day,
21:21 there's a fixed amount
21:25 of creative focus that I can
21:29 that I can engage with.
21:33 And I I I don't know dayto day how long
21:35 that's going to be. Some days I wake up
21:37 and it's six or seven or eight hours
21:40 where I can just focus and do like a
21:42 whole day of [ __ ] Some days it's like
21:45 20 minutes. And what I know is that and
21:49 and when I when I think about creative
21:55 mindset or create creative
22:00 engagement,
22:02 it doesn't mean I'm making pretty
22:04 pictures. For me, creativity is problem
22:07 solving.
22:09 And so, it's like, okay, I got this
22:10 thing I want to do. And so, it might be
22:11 I want to make a picture, I want to make
22:13 a video, or I want to write something or
22:14 I want to do this. And then it's like
22:16 it's all these minations about, okay,
22:18 how am I going to do that? What am I
22:19 going to do? What do I got to do?
22:22 And then on any given day, like the
22:24 amount of energy you have to do that is,
22:27 for me, it's fairly fixed. And if you
22:30 spend, let's let's say I woke up today
22:33 and I've got three hours in the tank.
22:37 If I spend those three hours
22:39 problemsolving
22:41 on something that's a low-level problem,
22:43 like how do I get this this this
22:47 word processor
22:49 to give me something formatted in a way
22:52 that I don't have to format each line?
22:54 Let me learn stylesheets and then I go
22:55 learn stylesheets. If I spend three
22:58 hours problemolving on low-level
23:02 tactical [ __ ]
23:04 then I have no energy left for the
23:07 higher level truly creative thinking.
23:12 And and what's struck me over the past
23:14 two years working with generative AI is
23:16 that
23:19 it's really good at some of the
23:22 low-level
23:26 blocking and tackling of creativity.
23:29 It's really good at it. And so very
23:32 often someone will come in here and say,
23:33 "Hey, how did well that this is what
23:35 happened the other night. How are people
23:37 making those Bigfoot videos I'm seeing
23:39 all over X?" Which I didn't know what
23:41 they were. We went and looked at some.
23:43 We said, "Ah, okay. I know how to do
23:44 that." And then I started to sit down
23:47 and say, "Okay, I want a video about a
23:49 Bigfoot." And then I was thinking about,
23:50 "Well, where do Bigfoots live? They live
23:52 in the Northwest and they Right. And so
23:54 we probably want something and we
23:56 probably we we want him looking at a
23:58 selfie. So like, how do you describe
24:01 selfie?" Like I I started going down the
24:03 rabbit hole of doing all the
24:05 problemolving
24:07 of figuring out what elements should be
24:09 in that prompt.
24:13 And then what hit me was this moment of
24:16 no, back out. Zoom back out.
24:20 Don't get seduced into using your brain
24:23 for low-level [ __ ]
24:27 because all of that stuff isn't really
24:29 creative. It's kind of tactical, right?
24:31 It's like, okay, what kind of camera
24:33 shot should it be? And what kind of
24:36 So, I zoomed out and then I said, "Hey,
24:38 chat GPT, there's these videos with
24:41 Bigfoots doing selfies. Can you write me
24:44 a a prompt for for a a a video tool that
24:48 does that?" And it wrote this. It wrote
24:50 what's on screen right here,
24:54 which is not super long, but it's it's
24:56 cool structure. It said, "Here's your
24:59 scene setup.
25:01 Bigfoot is holding a shaky selfie style,
25:03 a a shaky phone selfie style. He's
25:06 midbrant, sweaty, slightly out of breath
25:10 with a squirrel on his shoulder that he
25:12 ignores completely. Like like that
25:16 that line with a squirrel on his
25:18 shoulder that he ignores completely.
25:20 That's just good writing. That's just
25:22 good writing.
25:24 Right.
25:26 In fact, when we made this video,
25:28 someone in the comments said, um, "Is
25:31 that a squirrel on his shoulder?
25:35 Um, you get one continuous take. No
25:38 cuts, no effects, just raw Bigfoot
25:40 energy like that. That's just there's a
25:43 lot going on in here creatively that I
25:46 can look at and I can appreciate and and
25:48 I could do one of two things. I could
25:50 look at it and go, that's exactly what I
25:52 want." Or I could look at that and go,
25:53 "Oh, no. It completely missed the mark."
25:56 But see, now I'm just I'm operating at a
25:58 higher level. I'm operating at like the
26:00 producer level.
26:03 And so the the thing I want to start to
26:04 get my head around and I want to put
26:06 into this this new um project I have
26:10 called Feed the Prompt is
26:14 the ability to let AI
26:18 free you from the low-level crap that
26:22 chews up all your energy. So you can
26:24 just you can live at this higher plane
26:26 of creativity and just just be like
26:29 dictating down into the details. No, I
26:32 want it more like this. I want it more
26:33 like that. A little sprinkle of this. A
26:34 little sprinkle of that.
26:37 Um and then I went in and gave an acting
26:40 prompt. Look dead into the camera. Whyed
26:43 and unblinking with frantic intensity.
26:47 Deliver this line. I eat organic. I
26:49 compost my poop. And you all still think
26:51 I'm a myth. and it capitalized that.
26:54 This is a funny line,
26:57 right? I didn't have to come up with
26:59 that
27:01 archetypal architect. That's pretty much
27:02 how I go through go about writing.
27:06 I see and have a lot of images and
27:08 videos. They're cool, but I'm now stuck
27:10 on how to use these to make a difference
27:13 in my life with my work. Using them for
27:15 work sucks. It doesn't work.
27:17 Interesting. You know, Valerie, um,
27:21 Cindy [ __ ] sent me a I don't know if
27:23 Cindy's on the call tonight. It's Friday
27:25 night date night. She might be. Um,
27:27 Cindy sent me a video that she made
27:30 today in an hour in um in LTX Studio
27:38 and she's making things for work and
27:40 she's making things for clients and
27:42 she's doing them incredibly quickly. So,
27:47 so I think here's the trick, Valerie,
27:50 is
27:52 one is you got to play with the tools
27:53 enough to understand what's possible.
27:55 And I think you do that. You're here a
27:56 fair amount. But if you haven't played
27:58 with LTX Studio, there might be
27:59 something there. There might be also
28:00 something about connecting with Cindy
28:02 just to see see how she's doing it. I
28:04 actually suggested that Cindy do a learn
28:06 out loud on on this one project that she
28:09 showed me today. um because it's got a
28:12 lot of janky crap in it, but it doesn't
28:14 matter because it's it's just basically
28:15 this this I don't know. It it was it was
28:19 just this cool thing that communicated
28:21 an idea in video and it did it it did it
28:23 quite well and it didn't have to be
28:25 perfect. Um
28:29 but I think
28:31 I think the trick especially when you're
28:33 getting into video stuff, the the
28:35 problem with video is video is hard.
28:38 Video is complicated. It's structured.
28:41 It's layered. You've got to be good at,
28:44 you know, visual stuff and audio stuff
28:46 and musical stuff and sound effects
28:48 stuff and acting stuff and writing
28:51 stuff, right? And directing stuff and
28:53 and compositing it together stuff.
28:56 There's there's a lot of stuff you have
28:58 to do. So, it's very easy to to run down
29:00 rabbit holes with with video where
29:03 you're
29:06 letting the tool lead rather than the
29:08 idea lead. And I think I think the trick
29:10 with video is you got to be really clear
29:12 on what you're trying to accomplish for
29:14 what audience and then it gets a bit
29:16 easier. But but it's um
29:19 it's like that you're having trouble
29:22 with it at work doesn't surprise me
29:24 necessarily because if that's not what
29:27 you do regularly for work, it's going to
29:30 be this foreign territory. And I but but
29:33 again I think the key is start with what
29:34 are you trying to accomplish? Who's the
29:36 audience? Um I'll reach out. Yeah.
29:38 Great. Yeah, I would definitely connect
29:40 with her. Um,
29:43 okay. Archetypal Architect, you
29:45 definitely watch that, Lol. Yeah,
29:47 because I think Cindy, we we've done a
29:50 couple of, you know, things with LTX
29:54 Studio where we've gone in and said it
29:56 can do this and it can do that and it
29:58 can do that. But what I was thinking
30:00 about when Cindy sent me this one
30:03 specific output is I want her to do a
30:06 lull that starts by sh showing the
30:09 output and then go all the way back to
30:12 the beginning. What was she trying to
30:13 accomplish?
30:15 Who was the audience? Who was the
30:17 client? What was she trying to
30:19 accomplish? And then what was her
30:21 thought process? Did she start in chat
30:23 GPT? Did she go to Claude? Did she go to
30:25 Perplexity? Did she go straight into LTX
30:28 Studio? When she got in there, what did
30:29 she do? How did she pick a style? Right?
30:33 Like, if you want to knock something out
30:34 quick in LTX Studio, that flumxes me
30:38 because it's so complicated. There's so
30:40 many switches and dials. I get lost in
30:42 all of that. She's figured out a way to
30:44 navigate through that really quickly.
30:46 So, I think that would be quite
30:47 valuable.
30:49 Um, this would help me greatly in how I
30:52 think about video, I think. Yeah, like
30:54 like video again,
30:58 if you start with I need a video or I
31:01 need a video that's like this style and
31:04 then try to work your way back to the
31:06 story, it's really hard. You really have
31:09 to start with, you know, what am I
31:11 trying to accomplish? What's the story?
31:13 And then work forward to what's the
31:15 right tool because there's just too many
31:16 tools.
31:18 Um,
31:25 What did Leonardo put out today?
31:29 I use 4.1 for therapy and recipes. I
31:32 love that.
31:35 The New York Times and Gizmodo today.
31:37 Brandon sent this to me. Um had a thing
31:40 about how how AI is ruining people's
31:43 lives.
31:45 I wish VO3 had a prompting guide.
31:50 Yeah. Well, so
31:53 V3 probably does have a prompting guide,
31:57 but it's probably um it's probably
32:02 like
32:04 the history of cinema is the prompting
32:07 guide, right? Like, so in fact, let's
32:11 go. Let's let's Okay, let me go I'm
32:14 going to pull out here for a second. So,
32:17 I created a project called the video
32:19 scene builder.
32:21 And so I've got the Bigfoot thing we did
32:23 the other night is in here in here.
32:27 But let's start a new chat in here. And
32:29 let's say um
32:32 research
32:34 VO3 from Google
32:37 and tell me everything you know about
32:41 it.
32:44 um and recommend the best way ways to
32:52 prompt it to give
32:56 results that are professional
33:01 and
33:04 consistent across
33:08 scenes.
33:11 And then I'm going to turn on
33:17 Let's [ __ ] it. Let's do a deep research.
33:20 We'll we'll flip over to 03.
33:26 So So I'm in the model 03, which is one
33:28 of the reasoning engines. If If you're
33:30 How many folks are in here? 34. We
33:32 probably have some folks in here that
33:34 are new. If you're new to chat GPT,
33:38 GPT40
33:42 is a less sophisticated model than GPT
33:46 03.
33:50 Okay,
33:53 you're like, but why?
33:55 And then 04 mini is better than 03, but
33:59 it's mini, so it's not as good as 03. So
34:02 03 is the big model that does reasoning.
34:05 40 is the sort of daily driver. So we're
34:10 gonna we're gonna have it go do some
34:11 deep research and really think on the
34:13 best way. VO3
34:18 shock collar. Shut up.
34:21 03 Pro is going to blow minds. Yeah, I
34:24 don't have access to it yet. Okay. So
34:26 let's let's have this go do some work.
34:29 Could
34:30 you please clarify what you mean by V3?
34:34 Are you referring to specific video
34:35 generation model?
34:38 Yes, it's
34:40 VO3,
34:42 the new
34:45 video gen model used in flow.google.
34:55 All right. So, it's starting research.
35:02 capabilities, features, architecture,
35:04 best practices for prompting, generate
35:06 professional, consistent results. So,
35:08 here's here's what I'm here's what I'm
35:10 hoping we can do.
35:15 And I think you can now share
35:18 projects. So, this could be a project
35:21 that I I share on the on the salon.
35:28 This is I'm gonna need a drink type of
35:30 thing from chat GPT
35:35 starting research. Listen, I don't think
35:38 we need to do deep research on this
35:40 thing. Like I think we could have just
35:42 done a basic web search, but I always
35:43 think it's good to, you know,
35:46 use tools that are overkill for for the
35:49 for the challenge at hand.
35:52 Um, how did you find or get that scene
35:57 builder?
35:59 I just made a project called scene
36:01 builder. I that's that's I literally
36:03 just said like 10 minutes before I came
36:06 on the live, I had an idea for what to
36:08 do tonight. I thought so. One of the
36:10 things we talked about in office hours
36:13 and in the mastermind
36:16 um meeting today was was using projects
36:22 as a way to to
36:24 augment pro prompting. So here's here's
36:27 one of the things I learned today from
36:29 from the the sort of group conversations
36:31 that we had today. Um,
36:35 most people
36:37 that are writing sophisticated prompts
36:42 are having chat GPT or whatever the, you
36:46 know, their LLM of choice is helping
36:48 them write the prompts,
36:51 right? So, they're saying, "Hey, I need
36:52 a structured prompt that accomplishes
36:54 this or here's a big long prompt." Sam
36:57 Sam Elliott was talking about he got
37:00 hired into this job and he inherited
37:02 this 16page really sloppy messy
37:06 self-contradicting
37:08 prompt and he put it into chat GPT and
37:12 he said reorganize this remove anything
37:14 where it self-contradicts
37:16 and put it in structured organized
37:19 sections that make the most sense to
37:21 accomplish XYZ.
37:24 And now he's maintaining this much more
37:26 organized, you know, structured thing.
37:28 But how he's getting all of the
37:31 components and improving things is using
37:35 chat GPT. He said like if he wants it to
37:38 write a better email like one of the
37:40 sections is write good emails he'll
37:47 do a test where it says write an email
37:49 and then if he doesn't like the output
37:52 he'll copy the output paste it back into
37:54 chat GPT and say okay here's the prompt
37:57 here's the output it gave me and the you
37:59 know the prompt I gave it the output it
38:01 gave me and here's why I don't like it
38:04 rewrite this section ction of the big
38:07 prompt so that it improves that email.
38:11 And then he says it just will rewrite
38:13 stuff for him and then he just keeps
38:15 doing iterations like that. So
38:19 I I'm kind of I've never really prompted
38:22 that deeply
38:24 and I think it's a good skill to
38:26 understand and yeah it is this is
38:28 technically prompt engineering
38:31 but I think you can you can vibe prompt
38:35 engineer
38:39 which is if you know vibe coding right
38:41 you just ask for what you want and these
38:43 coding tools just code for you. We can
38:46 do the same with with prompting
38:47 sophisticated prompts. Do you think do
38:51 you have the project published yet?
38:54 Don't know how they work. Lol.
38:57 That's the first time. Yeah. Um I've
39:00 heard of it. Oh, okay. So, so let me
39:03 show you where projects are.
39:05 Projects are kind of cool. So, so you've
39:08 got custom GPTs, right? And you can make
39:10 one of those. projects are kind of like
39:14 a custom GPT, but they're kind of
39:16 different. So, let me go in. So, here's
39:18 one I've got for Sydney.
39:20 So, Sydney's my musical. So, you can
39:23 upload a project file, right? So, I've
39:26 uploaded the script. I've got custom
39:29 instructions. Act script and story
39:31 consultant for Broadway musicals. Don't
39:34 let us write crap. Everything has to be
39:36 award-winning. I don't know if that's a
39:37 good system prompt, but that's what it
39:39 is. And then here are all the different
39:41 prompts, all the different chat
39:43 conversations I've had about Sydney. So
39:46 now whenever I create a new chat inside
39:49 this project, it takes all of this other
39:52 stuff into into consideration.
39:55 The one that I just started for video
39:58 scene builder really has nothing in it,
40:00 right? It's got it's got the Bigfoot
40:03 comedy vlog prompt that we did and it's
40:07 got this research that we're doing right
40:08 now.
40:10 So that's it. And there's there's the
40:13 research it's doing.
40:15 But now what's cool is this whatever we
40:18 end up doing in this chat, this chat
40:21 itself now becomes source information
40:24 for the project.
40:26 Make sense?
40:28 So, think of a project. The project is
40:30 really just a folder with lots of chats
40:33 within it. But because you can give it
40:35 custom instructions and because you can
40:37 upload um data documents, source
40:40 documents,
40:42 it's kind of like um
40:47 like a like a less specific version of
40:50 Notebook LM. Notebook LM's like very
40:54 very specific to what it does. This is
40:56 sort of like that. It's sort of like a
40:58 custom GPT.
41:00 Um but but it's pretty cool stuff.
41:04 And it it just got more powerful. Like
41:06 they've added deep research. You can
41:08 have a conversation within it. You can
41:11 share them. You can do all sorts of
41:14 things. So, okay.
41:17 I've been having the AI prompt itself
41:19 since 3.0. Yeah, that's great. That's
41:21 awesome.
41:23 I posted it in the salon. Leonardo put
41:25 out a V3 prompting guide today. Oh,
41:27 cool. Whi which channel did you put put
41:29 it in, Vicki?
41:34 So, if you if you don't know what the
41:35 salon is, the AI salon, go to
41:41 go to that address, the salon.ai,
41:46 and click on join our community. And
41:48 that'll take you over to the community
41:49 site
41:51 and introduce yourself. join, jump in,
41:55 and then there's all sorts of different
41:57 spaces in there. Um,
42:03 [Music]
42:16 so
42:31 Um, where did Vicki put it? Tool talk.
42:37 All right. Tool talk. So, this is behind
42:39 the mastermind. It's in the mastermind
42:41 area. Beautiful. Um, so within the
42:44 salon, there's a free area of the salon
42:46 and then there's a subscription area
42:47 called mastermind that has some some
42:49 deeper level stuff in it. One of them is
42:51 tool talk. So, Leonardo just published a
42:53 guide for prompting in VO3. Very cool.
42:57 Mastering prompts for VO3. Let's go grab
43:00 this.
43:05 Planning your prompts. Okay, here's a
43:08 really good example
43:11 of
43:14 the the old way we used to do things
43:16 with our brains and the new way we're
43:18 going to do things with our with our
43:19 brains augmented by these tools. So,
43:22 somebody wrote this VO3 prompt
43:26 guide, right? Now, historically, what
43:29 you would do is you would print this
43:31 thing out, right? And then you would
43:33 have it beside you and you'd sit down to
43:34 write a prompt and then you go, "Okay, I
43:36 got to write a scene description. Then I
43:38 got to write a visual style. Then I got
43:40 to think about camera movement." And if
43:42 you don't know camera move lingo,
43:46 that would be a Google search to go
43:48 understand camera movement in Hollywood
43:50 and cinematography.
43:51 Then you've got to describe your main
43:53 subject. And is the main sub subject a
43:55 character or a car or a building or a
43:58 cloud? Ah, now like like at this point
44:01 I'm starting to get overwhelmed and
44:03 break into a flop sweat. And then you
44:05 got to get your background setting. And
44:07 then you got to get lighting right. And
44:09 then there might be an audio cue and a
44:11 color palette and dialogue and subtitles
44:13 and language, additional planning notes,
44:15 writing your prompt. Are you [ __ ]
44:17 kidding me?
44:19 So that's the old way. The old way is we
44:21 would get overwhelmed.
44:24 Oh, VO is on Leo. You can Google it if
44:26 you're not in the mastermind.
44:28 Yeah, Leonardo.
44:31 Prompts. You know, VO VO is on Leo.
44:34 That's funny. Check it out. Um,
44:38 but let's copy this.
44:41 Go back to chat. Jeep.
44:45 All right.
44:49 So, here's our deep research result.
44:56 This is big. Look at this.
44:58 I like it. Good lord. It's a lot.
45:05 Download PDF. Great. Perfect. Okay. So,
45:11 it it it researched for five minutes. It
45:14 found 19 sources and 34 searches.
45:18 This This is your brain. This is your
45:20 brain on AI. If you know, you know.
45:22 Actually, you know what, Kelly? I think
45:25 I'm going to steal that if you don't
45:26 mind. I'm going to put that in my in my
45:28 new presentation.
45:35 [Music]
45:40 So, we're going to do two things here.
45:47 Okay. Technical capabilities.
45:50 Great. Supported inputs, text, images,
45:53 and controls.
45:57 The primary input for V3 is a text
45:59 prompt,
46:01 but you can use images.
46:04 In Google's flow interface, it can
46:06 leverage ingredients,
46:09 dolly, zoom, aerial, wideshot,
46:11 generation, quality.
46:14 So, it's got some background [ __ ]
46:17 Camera and scene control scene builder
46:19 for continuous scenes outpainting. Okay,
46:22 great. So, we're going to download this
46:24 as a PDF.
46:32 All right. So, we now we now have the
46:34 deep research as a PDF. So, let's go
46:37 back out to the top level of our project
46:44 and we're going to upload
46:48 we're going to add files. So, we're
46:50 going to add
46:54 this thing that it just wrote. How many
46:55 pages is this? One, two, three, four,
46:58 five, six, seven,
47:02 eight. eight pages of of how to use V3
47:08 from deep research. We'll assume it's
47:10 good. It's probably not, but let's just
47:12 assume it's good.
47:15 All right. So, we have that in there
47:16 now. Um,
47:23 let's see.
47:27 I'm going to say based on the knowledge
47:34 You have now
47:37 and this article
47:40 about the best way to prompt
47:45 VO3,
47:50 there's the article. So, so I just
47:52 pasted in the article that someone just
47:55 g that Vicki gave us for VO3. Not
47:59 someone, Vicki. Vicki gave us it. I'm
48:02 bad at that. People give me stuff and
48:04 I'm like, a person somewhere in the
48:07 universe provided this URL. Thank you,
48:10 Vicki.
48:14 All right. Um,
48:19 comma, I want you to
48:24 create
48:30 a structured
48:38 prompt recipe.
48:45 Feel free to use JSo N JSON if you feel
48:52 that is important.
48:58 A structured prompt recipe and guide
49:07 that
49:15 captures
49:16 everything
49:19 a filmmaker
49:21 might
49:23 need.
49:34 to create both individual
49:38 clips
49:40 but also
49:42 consistent
49:46 um shots
49:52 throughout a story.
49:55 Okay.
49:57 So, so to be clear, and who was I think
50:01 it was
50:03 I think it was Kelly Camp. Yeah, it was
50:05 Kelly Camp today. This was in the
50:07 mastermind conversation.
50:10 Kelly Camp was sharing she uses chat GPT
50:14 to help her write prompts. But what she
50:16 does is she kind of vomits out all of
50:18 the kind of requirements for what she
50:21 wants in a prompt, which is that's
50:23 essentially what I'm doing here, right?
50:25 I just uploaded some knowledge, so use
50:27 that. Here's an article. Use that. And
50:30 now I want you to give me a recipe and I
50:32 want you to give me a guide. And then,
50:35 right, and so let's just let's just
50:36 generate this thing. And let's say we're
50:39 going to say um why don't
50:44 you put the resulting
50:49 guide in a canvas? And if you don't know
50:51 what a canvas is in chat GPT, it's a
50:55 separate editable document.
50:59 So if if you use chat GPT and it gives
51:01 you like some big long thing and then
51:03 you don't like one of the paragraphs
51:04 within it for the most part, you have to
51:07 like rewrite the whole thing. If you if
51:09 you put it in a canvas, you can just
51:11 rewrite little pieces of it.
51:16 All right. And so I forgot I was in 03,
51:18 but that's okay. We're going to do a
51:20 reasoning model version of it.
51:25 Oh, look. There's our our single JSON
51:28 template.
51:31 Good [ __ ] lord.
51:35 This is a lot. Like this is
51:37 overwhelming.
51:39 The anatomy of a V3 prompt.
51:43 V3 recipe and filmmaker guide.
51:47 Scene description. Scene underscore
51:49 description. So when I asked it to give
51:51 me a structured prompt, what it's doing
51:53 is it's writing all of the fields
51:57 that will be described in the JSON. So
51:59 JSON is just like a markup language.
52:02 It's JavaScript object notation I think
52:05 it stands for.
52:07 And large language models are very very
52:11 good at structured language, right?
52:13 Because a lot of what they teach these
52:15 things to do is code. So it's good at
52:17 structured language. So, it's defined
52:20 all of our different things. Color
52:22 palette, dominant hues, LUT, which is a
52:24 lookup table. Um, deep slate, gray,
52:27 ocean blue, and these are example
52:29 snippets.
52:31 All right. Enclose dialogue and straight
52:33 quotes if you want it spoken. Omit
52:36 quotes if you don't want subtitles. Tip.
52:38 Look, look at that tip right there. One
52:41 of the frustrations of um V3 is that it
52:45 adds subtitles and they're never right.
52:47 The words are always wrong, but it adds
52:49 them. If you put things in quotes,
52:51 you'll get subtitles. So, this included
52:54 this in our little thing here. So,
52:56 here's a single clip JSON template.
52:58 Scene ID, intro pool scene, scene
53:01 description, a glamorous woman with an
53:03 afro hairstyle and gold jewelry shares a
53:06 nighttime pool with a calm alligator.
53:08 All right, great visual life. Visual
53:11 style cinematic 35 millimeter film. So,
53:15 this is great. Okay. Field notes. All
53:17 right. Multi-seene continuity pack.
53:20 Great.
53:23 Iteration and troubleshooting. Um,
53:27 quick reference cheat sheet. Okay. So,
53:30 here's something that's missing from
53:32 here. when we did our um
53:37 our Bigfoot video clip the other night.
53:44 It gave us acting notes.
53:48 So,
53:50 so let's see. Um
53:53 the other night I did a Bigfoot
54:00 parody. Bigfoot selfie parody.
54:06 And it's pair OD, not pair A D, pair O
54:09 D.
54:12 And
54:14 the prompt
54:22 had
54:24 character and acting notes in it as well
54:30 as the technical
54:34 visual elements.
54:37 Um, your guide
54:40 seems to ignore
54:45 acting and character
54:52 all together.
54:56 Recommend how we incorporate
55:03 incorporate
55:08 this into
55:10 our guide.
55:14 So again, I could think through, well,
55:17 you could create it as a you could
55:19 create character as a separate module
55:21 and but [ __ ] it. Let it figure it out.
55:24 It's the reasoning model. Go off and
55:26 reason, chat GPT. Don't take up my
55:29 precious [ __ ] brain cycles.
55:31 [Laughter]
55:35 Kak Kyle, see my notes above about
55:38 Bigfoot. Okay.
55:41 [Music]
55:45 I'm laughing at these vids. They are
55:47 hysterical.
55:49 Here, I'll show you that one. Kak, there
55:51 is now a medium foot and a yeti
55:55 complaining about the beavers.
56:00 That's [ __ ] awesome.
56:02 Yeah. Like like I gotta tell you, man,
56:06 we're absolutely coming into this era of
56:10 joy where someone does something
56:13 creative and other people can
56:14 immediately respond to it, not with
56:17 shitty comments, but with their own
56:19 version of it and then like Yeah. Like
56:21 like Bigfoot and medium foot. That's
56:24 [ __ ] awesome. Okay. Anatomy of the
56:27 big. So did it. So let's see.
56:30 I expanded the guide to include
56:33 performance focused fields, performance
56:35 notes for emotion, gesture, comedic
56:38 beats, plus JSON keys,
56:43 continuity packed, iteration, and
56:45 troubleshooting. Chris reference cheat
56:47 sheet.
56:49 Um uh I think you also need to
56:57 include options for dialogue
57:03 in the absence
57:07 of dialogue.
57:11 Um,
57:13 VO3 will provide it, but it
57:19 would be nice to
57:23 leverage
57:26 powerful
57:28 LLMs to write
57:31 better dialogue
57:35 and have more control before
57:42 generating
57:45 the video
57:47 again. I'm going to say, "What do you
57:50 recommend?"
57:52 Let it do the [ __ ] thinking. I don't
57:54 need to use my brain to do your stupid
57:58 work.
58:02 What was the link for that research
58:04 thing, Vicki? I'm only finding the
58:06 Leonardo article on the salon.
58:12 I didn't I don't know which research
58:13 thing you were talking about.
58:20 IP for no one. Yeah, exactly. Well, Joy,
58:23 you know, it's funny, Joy.
58:26 So, IP for no one, intellectual uh
58:29 property rights for no one. Um I I
58:32 really do think we are entering a
58:34 postcopyright world. Not that you don't
58:37 want to protect your IP, but your
58:39 ability to protect it becomes
58:41 increasingly diluted. Um, and it's
58:44 already completely diluted, but it's
58:46 going to get ridiculous, right? And I
58:49 think what happens is like right now the
58:52 thing about copyright
58:55 is that
58:57 all of copyright law is about the
59:00 output,
59:02 right? Because when they wrote copyright
59:05 laws, it was actually difficult to copy
59:08 things.
59:10 So you'd produce a thing, an output
59:13 and then if someone copied it, that was
59:18 the producing of it was really valuable
59:20 and the the copies were hard and then
59:23 they got easier and easier and easier.
59:26 Now we're moving into a world where
59:28 there it's trivial like fan art is now
59:32 going to be as good or better than the
59:34 original work.
59:38 And so I think what has to happen is you
59:41 protect the copyright on the generation,
59:43 not on the output. You you do it on the
59:45 input, not the output. And that's where
59:47 you do attribution and that's where you
59:49 do compensation. I think that's the way
59:51 it's going to go. I think it's the way
59:53 it's got to go. But who knows? Who
59:55 [ __ ] knows? Okay. Dialogue options
59:57 are now fully woven into the guide.
59:59 There's a dedicated dialogue script with
1:00:03 key sub fields. Okay. Um, let's see.
1:00:06 Let's let's also add in also right now
1:00:10 VO3 clips are only 8 seconds long. So
1:00:18 this guide
1:00:21 should
1:00:25 um
1:00:29 guide the user
1:00:32 to create prompts
1:00:35 that are not that are um possible within
1:00:41 that constraint.
1:00:44 Constraint
1:00:47 All right, beautiful.
1:00:53 Oh, the deep research you saved as a
1:00:55 PDF. I was in and out of the room. I may
1:00:58 have missed a step. Oh, it's it's the
1:01:02 it's it's the one I saved. The the PDF I
1:01:04 saved. I haven't shared that yet. I
1:01:06 think what I'm going to do is I'm going
1:01:08 to share
1:01:10 I'll just rewind. Yeah, don't mind me.
1:01:12 I'm pretty high. That's okay. Listen,
1:01:16 you're high. I'm inarticulate, but
1:01:21 okay, great. Um,
1:01:25 there's a couple of things. I think we
1:01:27 can share projects now. So, I'm going to
1:01:29 share this project in tool talk in the
1:01:32 mastermind. So, anyone that's in
1:01:34 mastermind will have access to this
1:01:36 whole project.
1:01:38 um that specific thing, that specific
1:01:41 PDF that I did based on deep research. I
1:01:44 don't actually know if it's any good. I
1:01:46 I can maybe I'll I'll pop here's what
1:01:49 I'll do. I'll put I'll put the project
1:01:52 in mastermind tool talk tool tips. I'll
1:01:56 put that that PDF that I generated in
1:01:59 irregulars because I I haven't really
1:02:01 vetted it. Um I'm vetting this little
1:02:03 thing I'm making a bit more.
1:02:06 All right. dialogue script. Yo, humans,
1:02:09 just finished my morning jog. Feeling
1:02:11 legendary. Okay, great. So, that's good.
1:02:13 So, let's see.
1:02:16 So,
1:02:18 runtime reality check. V3 will not
1:02:21 generate clips longer than 8 seconds.
1:02:24 Plan your storytelling and tight beats,
1:02:26 one action at a time. Okay. Two. Single
1:02:30 clip JSON template. Same as before with
1:02:32 seconds. Wait, but it it ate all of our
1:02:37 Oh, okay. Wait. In the canvas,
1:02:42 you have now
1:02:45 truncated
1:02:48 the
1:02:51 JSON outputs
1:02:55 as
1:03:03 same as before,
1:03:07 etc.
1:03:14 return.
1:03:18 We need this canvas
1:03:22 to include the entire
1:03:28 prompting guide.
1:03:33 Please
1:03:34 return the detail.
1:03:38 You wait, you ninny,
1:03:41 [Laughter]
1:03:46 for people who missed office hours, the
1:03:48 summary is now on 100 flies on the wall.
1:03:50 Oh, that's awesome, Vicki. That's really
1:03:52 cool.
1:03:54 I'm excited to see it. All right, what's
1:03:55 going on over here? I so for the if
1:03:58 anyone's modding on Tik Tok, I know I'm
1:04:00 not paying attention right now and
1:04:02 there's a lot of comments in here. Um,
1:04:04 so I'm probably missing stuff.
1:04:12 Wait, I don't have the power anymore.
1:04:13 Hope he sees it.
1:04:16 If Is anyone modding? If is if anyone's
1:04:18 modding, pin something that I should be
1:04:20 looking at. And if not, I'll just scroll
1:04:23 back and we'll do this the oldfashioned
1:04:25 way. Remember when I used to scroll
1:04:26 through all the comments and I would
1:04:28 comment on them and people are like, "Is
1:04:29 he ever going to get to mine?"
1:04:33 I don't think we have any anyone modding
1:04:35 tonight.
1:04:37 Modding every Oh, Vickiy's modding
1:04:38 everywhere. Okay. All right. So, let's
1:04:41 see. The canvas now contains the full
1:04:43 unabbridged guide. Every JSON field,
1:04:45 continuity trick, dialogue workflow.
1:04:48 Okay. There's that. Leveraging LLMs for
1:04:52 tight dialogue.
1:04:56 Great.
1:05:00 Okay.
1:05:09 All right. So, now let's download this
1:05:11 as a PDF.
1:05:16 And then we're going to close this. And
1:05:17 we're going to go back out to our
1:05:19 project scene builder.
1:05:27 Did I do this as
1:05:30 Wait, where did
1:05:33 where did my that last conversation just
1:05:36 go?
1:05:49 Did we just [ __ ] lose a whole chat?
1:06:03 I think we did.
1:06:07 Huh? I can mod now. Okay. Kak new
1:06:11 subscription model. Is there
1:06:16 if there is content I used to see that I
1:06:18 no longer have access to?
1:06:25 You mean in in the mastermind and the
1:06:27 salon? Yeah. The the the clubs and um
1:06:33 some of the mastery workshop things like
1:06:36 tool tips and things like that are now
1:06:39 part of the mastermind that are behind
1:06:41 the payw wall. So, yes, some some stuff
1:06:43 isn't there. And then there's new things
1:06:44 that we're putting in there like
1:06:45 libraries and things like that.
1:06:52 All right. All right. So, I'm going to
1:06:54 add one more project here. Add files.
1:06:58 So, now we've got this prompting guide.
1:07:09 Okay. Boom.
1:07:12 Okay,
1:07:15 two files.
1:07:18 I don't know if I want custom
1:07:20 instructions yet. I'm kind of pissed
1:07:22 that I lost that last chat. Where did
1:07:24 that go?
1:07:26 Let me reload here. Maybe something's
1:07:28 funky with my browser.
1:07:35 Video scene builder. Ah, here we go.
1:07:38 Yeah, it was a screen redraw thing. All
1:07:40 right. So, here's my So, now I've still
1:07:42 got that prompt in there. That's or that
1:07:44 that chat in there. Okay. So, we now
1:07:46 have three different chats. We've got
1:07:48 our our Bigfoot chat.
1:07:51 We've got our
1:07:53 research that generated the the the VO
1:07:56 thing. And then we've got the the
1:07:58 prompting guide that incorporated that
1:08:01 PDF. And we created a prompting guide.
1:08:04 And then the prompting guide itself
1:08:05 became one of these files. So, we now
1:08:08 have two very focused files in here,
1:08:10 right? We've got the the V3 deep
1:08:13 research thing on prompting and we've
1:08:16 got our own generated
1:08:19 prompting guide that includes all the
1:08:21 video [ __ ] plus all the acting [ __ ]
1:08:24 Okay. So now
1:08:28 we should be able to
1:08:32 say,
1:08:34 I want you to create
1:08:41 a V3
1:08:44 prompt
1:08:52 for
1:08:54 Um,
1:08:56 the pilot
1:08:59 scene, the pilot 8second scene
1:09:09 of a new parody
1:09:14 of
1:09:18 Bigfoot
1:09:21 selfies.
1:09:26 called
1:09:30 microfut
1:09:35 selfies
1:09:37 where
1:09:41 microfut
1:09:43 beasts
1:09:47 are evil looking
1:09:54 but tiny
1:09:58 and have a short guy complex.
1:10:02 [Laughter]
1:10:08 Give me the first prompt
1:10:12 complete with funny dialogue to kick off
1:10:20 the series.
1:10:23 All right. So in theory,
1:10:27 in theory,
1:10:30 because we've done a project and because
1:10:33 we've got good high quality
1:10:36 guidance in it, what this thing's going
1:10:39 to produce is a structured probably JSON
1:10:45 prompt that has good comedy in it. Let's
1:10:49 Let's see what it does. Here we go. This
1:10:52 could be bad.
1:10:53 [Laughter]
1:10:58 Oh,
1:11:00 great. Google needs to hurry with the
1:11:02 extend feature, but I can wait. The
1:11:04 users creating to Okay, I've I
1:11:09 I should have turned off 03. So, by the
1:11:11 way, 03 is a reasoning model, so it does
1:11:14 thinking, so it's not as fast. So, if
1:11:16 you want to turn [ __ ] around quickly,
1:11:19 um, you should probably just flip it
1:11:21 back to 40.
1:11:25 All right, there's our prompt. Why it
1:11:27 works? It uses every field. Okay.
1:11:31 So, let's let's copy this
1:11:36 and let's flip over to this.
1:11:49 An Microfoot. An evil 6-in crypted
1:11:52 launches his first microfoot selfie
1:11:54 vlog, glaring into his phone with world
1:11:57 domination swagger.
1:11:59 Performance notes: Napoleon complex
1:12:01 strut, exaggerated arm swings, ominous
1:12:05 chuckle, halfbeat pulse before the
1:12:07 punchline. This is good dialogue,
1:12:10 script, microfut. Hey, giants, I'm 6 in
1:12:14 of pure doom. Kneel before me, redwoods.
1:12:17 Now hand me a ladder. This This lens
1:12:20 insults my stature. Okay. Delivery.
1:12:24 Raspy gremlin baritone. Cocky 110 per
1:12:27 minute sharp emphasis on doom and
1:12:30 ladder. All right.
1:12:34 Visual style handh held found footage
1:12:37 selfie horror comedy vibe. This is I
1:12:40 mean this is Blair Witch,
1:12:42 right? That just described Blair Witch
1:12:44 Project.
1:12:47 Handheld found footage selfie horror
1:12:50 comedy vibe. That's cool.
1:12:53 Arm length selfie closeup. Subtle bob as
1:12:57 creature hops up on a stump. Lens 14
1:13:01 millimeter fisheye.
1:13:04 So you see you see why
1:13:10 my resistance to use my brain is paying
1:13:13 off. This is a level of detail that I
1:13:16 just wouldn't I I just don't give enough
1:13:18 of a [ __ ]
1:13:20 Right. But if you want to get really
1:13:21 good results, you have to like be a pro.
1:13:24 Well, you don't have to be a pro. Let
1:13:26 the AI be the pro.
1:13:29 I watched a movie called Exists today
1:13:31 about Bigfoot. It's a found footage
1:13:33 movie. There you go. So, we could just
1:13:35 actually create one of those. All right,
1:13:36 let's go make let's go make this [ __ ]
1:13:39 All right, we're gonna we're gonna see
1:13:40 if this thing works.
1:13:43 Cam Ken, that's amazing. Isn't this
1:13:45 cool?
1:13:46 Right. So, so just to be clear, what we
1:13:49 started with was a prompt from two days
1:13:53 ago
1:13:55 where we had chat GPT take a basic idea
1:13:58 and make it better.
1:14:00 And then we said, "Hey,
1:14:03 use your deep research and go find out
1:14:05 everything you can about V3 and write me
1:14:08 a guide for for everything you know
1:14:10 about it, including how to prompt it."
1:14:12 And then Vicki found this Leonardo
1:14:15 thing.
1:14:16 And then we said,
1:14:19 "We want to we we want a guide to be
1:14:22 able to do structured prompts," which we
1:14:24 then modified because we had some
1:14:27 experience.
1:14:28 And so now we've got this project that
1:14:30 has all this intelligence and I'll just
1:14:32 be able to keep coming back in here
1:14:34 anytime I want to do something for V3
1:14:36 and throw it in here. Let's let's go see
1:14:38 if this actually works. Can you explain
1:14:40 the difference between 40 and 03? Yes.
1:14:44 Yes.
1:14:45 Okay.
1:14:51 Wait. Let's go. You know what we'll do?
1:14:52 Let me go. Let me jump over to flow.
1:14:55 Let's get this
1:14:57 rendering and then I'll explain the
1:14:59 difference between chat GBT 40 and chat
1:15:03 GBT3.
1:15:05 It is actually [ __ ] criminal at this
1:15:09 point that they're still allowing these
1:15:12 naming conventions to to exist to
1:15:15 coexist. It's really bad. Um, but that's
1:15:19 neither here nor there. We're going to
1:15:22 do text to video. One of the important
1:15:24 things with with flow is that this
1:15:26 always defaults to non
1:15:31 um
1:15:34 nonV3. It defaults to V2. So you got to
1:15:37 flip it to V3.
1:15:39 And then oh and by the way there's V3
1:15:42 fast and there's V3 quality.
1:15:48 All right. So V3 quality is 100 credits.
1:15:52 BO3 fast is 20 credits. So, I'm going to
1:15:54 go with that. Now, I'm going to paste
1:15:57 in. There's all my JSON.
1:16:00 You don't need to know what JSON is. You
1:16:02 don't need to know how to write it. It's
1:16:04 basically just tagged.
1:16:07 It's tagged text
1:16:09 is all you need to know about it because
1:16:11 it's our our little prompt machine is
1:16:13 going to write it for us. All right. So,
1:16:15 we're going to get that going. Okay.
1:16:17 So, while that's generating that, let me
1:16:19 just change back to this mode. Okay.
1:16:23 Because I got to talk with my hands.
1:16:26 Um,
1:16:32 a little history.
1:16:37 So, when when chat GPT came out, it was
1:16:40 Chat GPT
1:16:45 3 was what they called it. And then the
1:16:47 model sitting underneath it was GPT3 and
1:16:49 then it was GPT3.5
1:16:54 and then I think at some point
1:16:59 they called it chat GPT 3.5 and then
1:17:03 they called it chat GPT4.
1:17:06 So that so the early progressions was
1:17:09 the numbers were just sequential. It was
1:17:10 33 3.54
1:17:14 and then and and chat GPT4.
1:17:19 This this is probably a little bit too
1:17:21 much background, but I I hopefully it
1:17:25 will explain what's going on.
1:17:29 Chat PT4
1:17:31 was what they call a mixture of expert
1:17:33 models where it was actually 10
1:17:35 different
1:17:37 models that they they sort of added
1:17:40 together to create
1:17:43 um
1:17:46 to create a single model. So when you
1:17:48 would ask it a prompt it would say which
1:17:50 of these like one of them was good at
1:17:52 coding, one of them was good at writing,
1:17:53 one of them was good at whatever. There
1:17:55 were 10 different expert models that
1:17:56 that it it sort of dynamically stitched
1:17:58 together. The reason that's relevant is
1:18:01 it was a 1.8 trillion
1:18:05 parameter model. It was 10 whatever it
1:18:08 is 180 billion parameter models stitched
1:18:12 together. It was very big and very
1:18:15 expensive.
1:18:17 And then they came out with chatpt 4 and
1:18:20 the O stood for omni. And omni means
1:18:23 multimodal. So it could do images, it
1:18:26 could see images,
1:18:28 uh it could do the voice thing and it
1:18:32 was much smaller. So they were getting a
1:18:35 similar quality of output with
1:18:38 multimodal capability
1:18:42 um
1:18:47 for for way cheaper and that became the
1:18:49 default model. So for a long time there
1:18:51 was GPT40
1:18:53 which was this tight omni model and then
1:18:58 sitting underneath it was GPT4
1:19:01 which was this big ass behemoth thing.
1:19:04 They've since retire retired GPT4 I
1:19:07 think. Let me go just make sure it's not
1:19:09 still there.
1:19:11 Yeah, it's it's gone. So So four is
1:19:14 gone. 40 was is is kind of the daily
1:19:17 driver model.
1:19:21 03.
1:19:23 This is just [ __ ] bonkers.
1:19:31 They have 40
1:19:33 and then they said, "Hey, we figured out
1:19:35 a a new way
1:19:38 for this large language model to
1:19:40 behave."
1:19:42 So they took sort of a base model and
1:19:45 then rather it
1:19:48 giving you a single result for a single
1:19:51 input
1:19:53 they created this model called 01 not 10
1:19:58 O 01.
1:19:59 So 40 the O stands for omni
1:20:04 01 the O stands for open AI1.
1:20:08 It was the 01 reasoning model.
1:20:13 And what the 01 model did is when you
1:20:16 would give it a prompt, it would it
1:20:18 would prompt and answer itself multiple
1:20:21 times before it actually answered you.
1:20:24 It's what they call reasoning. It's not
1:20:26 actually reasoning from a definitional
1:20:31 uh point of view, but but the all of the
1:20:33 models that have O in front of them are
1:20:36 reasoning models. Okay. So, so we had 01
1:20:41 and then you had 01 mini and 01 mini
1:20:44 high. So, 01 is the big model and the
1:20:47 mini models are smaller and not as
1:20:50 expensive.
1:20:53 They skipped O2 because someone else had
1:20:56 the trademark. So, 03 is actually O2.
1:21:02 But it's the O in front of it that lets
1:21:04 you know it's a reasoning model. And the
1:21:06 40 is an omni model. Here, let me show
1:21:09 you one other thing. Let me flip back
1:21:11 and share share my screen.
1:21:16 Um,
1:21:19 so, so you've got 4 O. The O after the
1:21:22 number means it's it's an it's a
1:21:26 multimodal model that can understand
1:21:30 images, generate images, understand
1:21:32 audio, generate audio, generate text,
1:21:35 generate code, all that sort of stuff,
1:21:37 understand all that, do reasoning, write
1:21:39 code, right? So it's it's a big ass
1:21:41 fancy model. 03 is the one that when you
1:21:45 prompt it, it says thinking and then it
1:21:47 says, "Okay, the user wants to do this.
1:21:49 Now I'm going to go do some research.
1:21:51 Now I found this. Now I'm going to tell
1:21:52 the user this. So it does all this back
1:21:54 and forth stuff.
1:21:57 04 Mini and 04 Mini High are are the
1:22:01 small versions of 03 or or of the next
1:22:05 model beyond 03 that we're getting kind
1:22:08 of a a sneak peek into what's coming.
1:22:12 And then down here you've got GPT 4.5
1:22:15 and GPT 4.1 and 4.1 mini.
1:22:19 None of these are are multimodal models,
1:22:22 but they are sort of the next just
1:22:24 different flavor. It doesn't [ __ ]
1:22:26 matter.
1:22:28 The important thing to know is just
1:22:30 these two. GPT40, that's your daily
1:22:33 driver. 03 is the reasoning model. And
1:22:35 then if if you want to get more
1:22:36 sophisticated, you can play with the
1:22:38 other ones. But just know that those two
1:22:39 things are different. If you want a fast
1:22:41 response, choose 40. If you want a a
1:22:44 thinking response, choose 03. And 03, to
1:22:48 be clear, is a more sophisticated, not a
1:22:51 more sophisticated model. It's a very
1:22:53 different kind of
1:22:56 interaction than you get with 40. So
1:22:59 sometimes 03 is really good for
1:23:00 something. If you just want like a quick
1:23:02 headline or something like that, 40 is
1:23:04 way better.
1:23:06 If you wanted to do stuff like go
1:23:07 research everything you can on VO and
1:23:09 write me a report. That's why I went to
1:23:11 to 03 because it's using the reasoning.
1:23:15 All right. Make sense? I know it's it's
1:23:18 it's [ __ ] insanity. So is so so wait.
1:23:21 So is 04 mini this is great. My 40 has
1:23:25 started thinking. So two block two block
1:23:26 Tom the 40 starting to think there's
1:23:31 speculation that they're sneaking some
1:23:33 of what's coming in GPT5 into 40 right
1:23:37 now. So if your 40 says thinking they're
1:23:40 starting to put some reasoning into the
1:23:43 the the base model.
1:23:46 What they have promised us and I I I
1:23:49 give this zero [ __ ] credibility at
1:23:52 this point. What they have promised us
1:23:54 is that GPT5 will get rid of this
1:23:57 nonsense
1:23:59 that GPT5 will be a single model that's
1:24:02 got an orchestration layer that that it
1:24:04 figures out which model it should use
1:24:06 underneath
1:24:09 Cam Katkin. So is 04 mini better than 03
1:24:13 or is it just faster? Okay,
1:24:16 04 mini high
1:24:21 some people say is better than 03.
1:24:27 So, so it is definitely the next
1:24:32 base reasoning model, but it's the mini
1:24:34 version of it. So if you think 03 I I
1:24:36 don't know what their parameter numbers
1:24:38 are at this point but let's just say
1:24:40 let's just say 03 is you know 500
1:24:44 billion parameters 04 mini is probably
1:24:48 you know a 100red billion parameters or
1:24:50 or 50 billion parameters right it's it's
1:24:53 a way smaller model but it is the next
1:24:56 it is the next model right so apparently
1:25:00 04 mini high is better at coding than 03
1:25:04 3, but 03 is also really good at coding.
1:25:07 So, I think this is one of those things
1:25:09 where
1:25:12 experiment with it. If if you're not
1:25:14 getting the result out of 03, try 04
1:25:17 mini. Try 04 mini high. Now, they also
1:25:20 did just come out with 03 Pro, which is
1:25:23 only available on the pro account. And
1:25:26 that's one where I think what makes it
1:25:28 pro is they just run it longer.
1:25:33 So, so when you put a prompt in 03 Pro,
1:25:37 it runs for like 15, 20, 30 minutes
1:25:40 before it gives you a result. Like 03
1:25:43 will run for five minutes and then give
1:25:46 you a result. 03 Pro is just running
1:25:49 much longer. So, but apparently 03 Pro
1:25:52 is delivering really high quality
1:25:54 results. So, good flipping Lord. It's,
1:25:58 you know, I'd forgotten about All about
1:26:00 01. I know. Well, and now like now that
1:26:03 you've got 40 and 03 03, you know,
1:26:09 people think that three is worse than
1:26:12 four because that's how software's
1:26:15 always [ __ ] worked. You you always go
1:26:18 by the number.
1:26:20 if they had called it like
1:26:23 GPT reasoning 03 or GPT reasoning 3 and
1:26:28 then just you know um GPT for
1:26:33 multimodal or something like that like
1:26:35 if if they just given them some name
1:26:39 that says this is a different thing I
1:26:41 don't think people would be confused
1:26:44 like what the [ __ ]
1:26:47 can no one does does no one there have a
1:26:50 liberal arts degree that can say this is
1:26:53 [ __ ] stupid.
1:26:57 I mean, it's staggering to me. They're a
1:27:00 [ __ ] what? $3 trillion company at
1:27:02 this point. They can't hire a a branding
1:27:06 consultant for a week. I'll do I'll
1:27:10 we'll [ __ ] do it right now.
1:27:16 03 Pro.
1:27:18 Will 03 Pro run scenario situations?
1:27:22 Probably. So like 03 03 Pro is when when
1:27:26 you want to solve big ass problems.
1:27:31 So I would think so
1:27:33 because one of the things that 03 has
1:27:36 that is mindblowing
1:27:39 is it can use tools. And not only can it
1:27:42 use tools, it writes its own tools.
1:27:46 So if if in its thinking it realizes it
1:27:49 needs to do some statistical analysis on
1:27:52 some data it found it will write and
1:27:54 execute Python code to do the
1:27:56 statistical analysis as part of the
1:27:59 reasoning.
1:28:00 So I would think that 03 Pro does that
1:28:03 on steroids, right? It's probably crazy.
1:28:07 Honestly, I think I prefer this to
1:28:09 something like how they name car models
1:28:11 or engineering model numbers.
1:28:14 you prefer what to it? What they've done
1:28:16 because I think what they've done is how
1:28:18 they're naming car numbers, right? It
1:28:21 does seem like they could have just
1:28:22 followed the frigin software conventions
1:28:25 or just giving them a [ __ ] name like
1:28:26 like the like GPT
1:28:29 reasoning
1:28:32 three.
1:28:34 You've got GPT4 and GPT reasoning 3.
1:28:37 That would solve it. Well, it wouldn't
1:28:39 solve it, but it would at least be
1:28:40 clear.
1:28:44 You know, you have GPT4 with pictures,
1:28:49 GPT reasoning 3, GPT 4.5 creative
1:28:54 writing, GPT 4.1 fancier but not quite
1:28:58 as cool as GPT4 pictures. Like,
1:29:04 just, you know, give a dude a break. All
1:29:07 right, let's go see our video.
1:29:13 Oh, that's cool. Remember we we said it
1:29:15 was a like a What did it say? That it
1:29:18 was a fisheye lens.
1:29:21 10 mm fisheye lens.
1:29:35 Oh [ __ ] We got um we got subtitles.
1:29:39 A giants. I'm 6 in of pure doom. Kneel
1:29:42 before me, Redwoods. Now hand me a
1:29:44 ladder. This lens insults my stature. A
1:29:47 giants. I'm 6 in of pure doom. Kneel
1:29:50 before me. Okay. So, let's go back to
1:29:53 our chat jpai and say um I'm gonna say
1:29:56 listen,
1:29:59 you had the rule
1:30:02 that if you include
1:30:09 quote marks, it would generate
1:30:15 subtitles.
1:30:17 It did. And I don't want subtitles.
1:30:23 Also, the
1:30:26 microfut
1:30:34 is holding a phone.
1:30:38 we can see and it should look more like
1:30:45 he is
1:30:47 filming with a phone we can't see.
1:30:55 Update the prompt,
1:30:58 you ninny.
1:31:02 Oh [ __ ] I'm not sharing my screen. Shut
1:31:04 up. Shut up.
1:31:07 All
1:31:14 right. What did it say? It didn't say
1:31:16 anything. No subtitles. No visible
1:31:19 phone. Copy. Go back over to flow. Flu.
1:31:24 Hello. Hello. And flu. Flu.
1:31:29 Bang.
1:31:36 All right. So, that thing's generating.
1:31:39 All right. We'll wait for that.
1:31:42 Um,
1:31:46 use the selfie stick stick in the
1:31:48 prompt. Yeah, that's pretty good idea.
1:31:52 All right, beautiful. All right, let me
1:31:54 go back and look at Tik Tok comments
1:31:56 because there's a lot of stuff going on
1:31:57 in Tik Tok. Tik Tok's being really janky
1:32:00 for me. So, night all. Uh, all right.
1:32:04 Bye, Silver Fox.
1:32:06 Sorry, it's being janky.
1:32:13 4.5 is creative writing. Yeah. So, 4.5
1:32:18 is
1:32:21 is not multimodal, so it doesn't do the
1:32:23 image generation stuff.
1:32:26 The rumor that I've heard is that 4.5 is
1:32:29 actually going to be the primary
1:32:33 mediating layer of GPT5
1:32:37 that it's it's really good at empathy
1:32:40 and understanding instructions and it's
1:32:42 it's apparently quite good at creative
1:32:44 writing. So, if you're having not good
1:32:48 results with 40, you should absolutely
1:32:50 try 4.5 for creative writing. It's
1:32:53 probably worth experimenting with just
1:32:54 to see what it does better. But yeah,
1:32:56 it's a creative writing model.
1:33:03 We need an open AI model bingo card. We
1:33:05 really do. It's so bad.
1:33:10 Looking to get into a rabbit hole
1:33:11 tonight with Zapier. That's fun. Can't
1:33:14 they use chat GPT to design a branding
1:33:16 campaign? Seriously, I'm a librarian.
1:33:19 They they could they could [ __ ] knock
1:33:21 this out in an hour. Like we could do it
1:33:24 right now.
1:33:27 See, this still gave us um
1:33:52 All right, we got to figure out a way to
1:33:54 fix this dialogue thing. So, let's go
1:33:58 look at our JSON. We'll find our
1:34:02 dialogue dialogue
1:34:05 script. All right, let's get rid of the
1:34:06 prompts. Hey, Giants.
1:34:13 All right, let's just try that.
1:34:18 Um,
1:34:20 go.
1:34:24 We need an open AI model bingo card. All
1:34:26 right, let's see what else we got in
1:34:27 here. Um, if they only had a reasoning
1:34:31 model to help with naming conventions.
1:34:32 Exactly, Sasha. It's like, seriously,
1:34:35 what is going on?
1:34:38 Ask Gemini. Yeah.
1:34:42 Uh
1:34:49 you can get V3 for free so long as you
1:34:52 as you have Google Workspace. You you
1:34:54 don't get that many generations.
1:34:58 They don't have qualifications.
1:35:04 Sahil, can I ask you a question? Yes.
1:35:07 Ask away.
1:35:10 Did you already ask a question? I'm
1:35:13 because I was focused on building us a
1:35:17 thing.
1:35:19 I didn't see your question if you asked
1:35:20 it, but yeah, ask away.
1:35:26 Horrible Park.
1:35:35 All right, let's see.
1:35:38 Can we start a change.org petition to
1:35:41 take over AI open AI naming? I don't
1:35:44 think I I apparently with GPT5 we're not
1:35:49 going to need it because we we won't
1:35:52 won't have to deal with the models
1:35:53 anymore. But here's the problem. This is
1:35:56 a problem with all software. They're
1:35:59 like, "Hey, listen. GPT's 5 is coming
1:36:01 out in a month or two. So, let's we'll
1:36:04 just keep the names the way they are and
1:36:06 and it it won't be that long because we
1:36:08 we'll have GPT5 out by the end of April
1:36:10 or the beginning of May. Well, now we're
1:36:13 in the middle of June and GPT5 is not
1:36:16 coming out I think until July and that
1:36:18 may push till August.
1:36:21 And then since then they've launched GPT
1:36:25 4.5, 4.1, 4.1 high, 04 mini, 04 mini
1:36:31 high on top of 03.
1:36:36 For a while they they they had 03 mini,
1:36:38 03 mini high in there. And for a while
1:36:41 they had 03 mini high with 01 and 01
1:36:45 pro.
1:36:47 So, in the past
1:36:49 three, four months, we've had 10 models
1:36:53 on top of the core model and the one
1:36:57 below it, the 03 model. Won't be that
1:37:00 long. And yet, we'll have five more
1:37:01 model names. I know, I know, I know, I
1:37:04 know. This is why they need a cranky Gen
1:37:07 Xer in the corner office at OpenAI. Hey,
1:37:09 Sam, I'm here. I know you're jealous of
1:37:12 me. Just put me down the hall.
1:37:15 Uh, I'll come scream at you
1:37:17 occasionally. Hey, Sam. Uh, [ __ ]
1:37:20 knock it off with the names now. Well,
1:37:23 we got 05. Yeah, but it's never gonna
1:37:25 launch. So, could we just [ __ ] fix it
1:37:27 short term? Just rename them so that
1:37:29 people aren't losing their [ __ ] minds
1:37:31 and not using the right models.
1:37:38 I'll be cranky in the corner for free. I
1:37:41 got my family on Grock because chat GPT
1:37:43 is so confusing and complicated with all
1:37:45 the models. Exactly, Robbie. Exactly.
1:37:48 Like there, you know, they got to get
1:37:50 their [ __ ] together. It's like needing
1:37:53 to have a dozen open tabs. They won't be
1:37:55 able to hire just one model. Well, in
1:37:58 theory, the the new model will do all
1:38:00 the all the routing. And but you know
1:38:03 what they found out?
1:38:06 why chat GPT hasn't launched or GPT5
1:38:09 hasn't launched yet.
1:38:11 They found out that even chat GPT can't
1:38:14 figure out what [ __ ] models to use in
1:38:16 what situation.
1:38:19 So like they're saying to us, here's a
1:38:21 consumer app, chatgpt, right? You go to
1:38:24 chat.com
1:38:25 and it takes you to chat GPT. That's a
1:38:28 consumer play.
1:38:31 The thing that made Chat GPT compelling
1:38:33 when it came out was you didn't have to
1:38:35 pick models. You didn't have to do the
1:38:38 temperature dial. You didn't have to
1:38:39 understand what coding was and what a
1:38:41 system prompt was. You just stuck in
1:38:43 words and out came an answer. And it was
1:38:46 [ __ ] brilliant. They got to 100
1:38:47 million users in six weeks. And now
1:38:51 we've got eight [ __ ] models to choose
1:38:53 from.
1:38:55 They don't understand who their customer
1:38:57 is. Like it's just it's [ __ ]
1:38:58 insanity.
1:39:00 [ __ ] insanity. All right.
1:39:05 Good lord. All right.
1:39:08 So, it will be like a switchboard. It's
1:39:10 literally going to be like a
1:39:11 switchboard, Gavin. It's It's going to
1:39:13 be one ringy dingy, one ringy dingy.
1:39:16 Yep. It's going to be Lily Tomlin at the
1:39:18 switchboard plugging in wires. So, when
1:39:21 we type something into chat GBT, it's
1:39:23 going to go like, "Hm, I wonder which
1:39:26 model would be best for this." I know.
1:39:28 We'll use a combination of two models.
1:39:31 Click. Click.
1:39:35 I hope. I hope it does that. All right.
1:39:39 Hang on. I got to change my audio
1:39:40 settings or you won't be able to hear.
1:39:42 You'll be like, "Kyle, we can't hear the
1:39:44 little gremlin. We want to hear your
1:39:46 microphone. We can't hear it. What's
1:39:48 going on? This is not very fair to us
1:39:50 people on YouTube." All right,
1:39:53 here we go. It got the phone in the
1:39:56 frame, which I don't like, but whatever.
1:39:57 Giants. I'm 6 in of pure doom. Kneel
1:40:00 before me, Redwoods she. Now hand me a
1:40:02 ladder. This lens insults my stature.
1:40:05 Eight giants. I'm 6 in of pure doom.
1:40:08 Kneel before me, Redwoods sheep. Now
1:40:10 hand me a ladder. This lens insults my
1:40:12 stature. Giants. You know what? I want
1:40:15 to try that same prompt with the higher
1:40:19 quality model.
1:40:22 Um, so I think I do that. Yep, there we
1:40:25 go. But then I'm going to do
1:40:28 higher quality model.
1:40:32 We'll we'll spend uh we'll spend 100
1:40:34 credits on this one.
1:40:36 All right, there we go.
1:40:40 I forgot how to use a GPT on the chat.
1:40:43 Wait. Oh, can you please show us how to
1:40:47 use Canva GPT?
1:40:51 Do you mean
1:40:53 do you mean you mean canvas within chat
1:40:57 GPT
1:41:00 cosmic?
1:41:06 Did you already talk about lovable being
1:41:08 free this weekend? I didn't Rick Olsen.
1:41:10 That's a good reminder. Thank you. So
1:41:13 all weekend this weekend lovable.dev is
1:41:16 free. you can do. They might rate limit
1:41:18 it if their servers start melting down,
1:41:22 but they're doing a contest this weekend
1:41:23 and they just kicked off this thing
1:41:25 called Lovable Shipped, which is a I
1:41:28 think it's a six-w weekek accelerator
1:41:30 program for people that applied to it
1:41:32 that said, I want to do cool [ __ ] with
1:41:34 Lovable. So, that kicked off, but
1:41:36 they're kicking it off with it's free
1:41:38 this weekend. So, if you haven't tried
1:41:40 vibe coding, if you don't know what vibe
1:41:42 coding is, vibe coding is basically
1:41:45 where you just speak a web app into
1:41:49 existence or any app really depending on
1:41:52 the tool. Lovable just produces web
1:41:54 apps. So, so web applications,
1:41:58 but you can literally just say, "Make me
1:42:00 a clone of the game Asteroids," and it
1:42:03 will make that or make me a recipe
1:42:05 database that people can log in and put
1:42:08 in their family recipes and it will go
1:42:11 build you the entire app and then you it
1:42:13 builds it and then you're like, I don't
1:42:15 like this about it and it'll go fix
1:42:16 that. So, it's free all weekend. If
1:42:19 you've never tried it, it's [ __ ]
1:42:22 remarkable.
1:42:24 If you've never tried vibe coding, it's
1:42:26 absolutely insane. GPTs, any GPT like
1:42:29 the Sin City Sipper. Okay, Cosmic Lover,
1:42:31 I'll show you how to do that. Um, let me
1:42:34 let me let this uh well, no, let me let
1:42:37 this video render and then I'll hop over
1:42:39 and uh and show you quickly how to how
1:42:42 to create your own custom GPT. In two
1:42:45 years, I'm going to vibe code a
1:42:46 smartphone operating system. Yeah. Well,
1:42:50 I you know what's funny, Robbie? I think
1:42:52 like by the time we get to two years, I
1:42:55 think software development is going to
1:42:57 be an entirely personalized dynamic
1:43:01 thing. I think software as we know it
1:43:05 will effectively cease to exist. It'll
1:43:07 it'll sort of atrophy away. And what's
1:43:10 going to happen is
1:43:12 as we need software,
1:43:14 we will sort of birth it into existence.
1:43:17 It'll do what we need it to do and then
1:43:18 it'll go away. Um there may be
1:43:22 persistent things like data but I think
1:43:24 even even things like databases like a
1:43:27 CRM I could imagine that CRM dynamically
1:43:31 reconfigure themselves. There's a core
1:43:33 data set and then how they behave is
1:43:37 dependent on what you need in the
1:43:39 moment. I like I think we go to an
1:43:41 incredibly personalized place. Source
1:43:44 camp, thank you. I haven't really needed
1:43:46 Claude much lately. I'll give it a try.
1:43:48 We might not have smartphones in two
1:43:50 years. That's true. We might not. What's
1:43:53 free all weekend? Lovable.dev.
1:43:56 Lovable.dev.
1:44:04 Just need a lot more hardware and
1:44:07 infinite energy. Yeah, exactly. So,
1:44:11 so, so Pate who's in the in the chat,
1:44:14 you know, works for Google on on
1:44:17 optimizing Tensor there. Oh, man. This
1:44:20 added flipping.
1:44:26 Dang it.
1:44:29 Dialogue.
1:44:32 Let me generate another one. It added
1:44:34 subtitles to it.
1:44:41 Eight giants. I'm 6 in of pure doom.
1:44:44 Kneel before me, Redwoods G. Now hand me
1:44:47 a ladder. This lens insults my stature.
1:44:49 Eight giants. I'm 6 in of pure doom.
1:44:53 Kneel before me, Redwoods G. Now hand me
1:44:55 a ladder. This lens insults my stature.
1:44:58 Eight giants. I'm 6 in. You know what's
1:45:00 cool though with this script? Because
1:45:03 it's got all that detail in it. Like the
1:45:07 acting here is really consistent because
1:45:08 the acting notes are it it's hitting the
1:45:11 right. It said hit the words ladder and
1:45:14 doom and it it did which is amazing.
1:45:18 It must be in training.
1:45:28 What software are you using for that
1:45:30 video? This is um this is Google's V3.
1:45:35 Um, I'm at flow.google. flow.google is
1:45:39 so um, so Google partnered with Darren
1:45:43 Arnowski, a filmmaker, and a bunch of
1:45:45 independent filmmakers. So when they
1:45:47 created the V3 model, they they they
1:45:50 brought in folks from Hollywood that,
1:45:51 you know, didn't want to burn the
1:45:53 building down
1:45:55 because some of some of the Hollywood
1:45:57 people are pretty pissed right now. But
1:45:59 some of them are like, oh, like James
1:46:01 Cameron's like, oh, here's a new tool
1:46:03 for self-expression. What's most
1:46:05 important is the story. It doesn't
1:46:06 [ __ ] matter what tool you use, right?
1:46:08 So, so some of them get it. So, Darren
1:46:12 Arnav Ornowski, I think is how you say
1:46:14 his name, um, is one of those
1:46:16 filmmakers. And then they brought in a
1:46:18 bunch of independent filmmakers.
1:46:20 Flow.google is a really hyper
1:46:24 lightweight scene editor where I can
1:46:28 generate these clips. And you'll notice
1:46:29 up here in the lefthand corner of each
1:46:31 clip, it says add to a scene. I can
1:46:34 stitch these things together and make a
1:46:36 little movie. Um, it it doesn't really
1:46:38 have much in the way of sophisticated
1:46:41 timeline editor stuff. It was pretty
1:46:43 cool. Um,
1:46:46 all right. So, let's see. That thing's
1:46:48 still going. Let me just show you
1:46:49 quickly. Um, Cosmic, here's So, if I go
1:46:53 to a new chat. So, here. Oh, wait. I got
1:46:56 to change my sharing. See, this is why I
1:46:59 didn't want to flip out of here because
1:47:02 everything's a [ __ ] production. All
1:47:04 right. Um, so here you are at chat GPT.
1:47:09 On the left hand side, you've got, you
1:47:11 know, your your basic elements at the
1:47:13 top. You have whatever you have access
1:47:15 to like code, codeex, Sora, and then
1:47:17 you've got your custom GPTs, right? So,
1:47:20 here's my custom GPTs including there's
1:47:22 Sin City Sipper.
1:47:25 Um
1:47:28 and let's see
1:47:34 GPTs. Yeah. Okay. So, you click on GPTs
1:47:39 and then that gives you the big long
1:47:40 list. This is the the
1:47:43 chat GPT store, right? So, you can find
1:47:46 other people's GPTs. You can search here
1:47:48 but in the upper right hand corner you
1:47:51 click on create GPT
1:47:54 and that'll take you to this dual window
1:47:56 model where you can just say I want a
1:47:58 GPT that you know um help helps me
1:48:02 calculate
1:48:04 when the next full moon is and it will
1:48:06 it will do something and create it here
1:48:09 and then over here in the right you can
1:48:10 play with it and then so you can either
1:48:13 do this in create mode where you're
1:48:15 you're literally just talking to it like
1:48:17 like a chat GPT chat and it'll build you
1:48:20 the custom GPT or you can go into
1:48:23 configure mode where you upload a
1:48:25 picture, give it a name, a description,
1:48:27 here's your prompt, um here are the
1:48:31 different tools it has access to.
1:48:34 You can upload data files to it, things
1:48:36 like that. So that's in the configure
1:48:38 tab. So that's it. And it's like you can
1:48:40 you can create a custom GPT literally
1:48:43 within minutes. So, you just go to GPTs,
1:48:46 the little GPTs thing on the left, and
1:48:49 then in the upper right hand corner
1:48:51 right here, you say create, and then
1:48:53 you're off to the races.
1:48:55 All right. All righty.
1:48:58 James Cameron doesn't need actors
1:49:00 anymore. I know, but neither do any of
1:49:02 us.
1:49:06 This is a democratizing tool. Listen,
1:49:09 just like it used to be, if you wanted
1:49:11 to shoot a film, you had to have enough
1:49:14 money to go buy canisters of 35
1:49:18 millimeter film and cameras that could
1:49:20 hold that film and operators that knew
1:49:23 how to work that film. And then you had
1:49:24 to have enough money to have that film
1:49:26 developed. And then you had enough money
1:49:28 to have an editor who had a big giant
1:49:30 machine to slice together 35 millimeter
1:49:33 film and track the audio and do the and
1:49:35 then and then they came out with Final
1:49:37 Cut Pro.
1:49:39 And they're like, "Oh, that's going to
1:49:40 ruin the film industry because now any
1:49:43 Yahoo with a [ __ ] Mac
1:49:45 can make a movie. What about us?"
1:49:50 Same [ __ ]
1:49:52 same panic.
1:49:55 Storytelling doesn't change.
1:49:58 The tools are changing. How you produce
1:50:00 [ __ ] changes. It just happens to make it
1:50:03 more accessible to literally millions of
1:50:06 people.
1:50:08 But it's still about telling compelling
1:50:10 stories. And if you can tell a
1:50:12 compelling story, whether you use film
1:50:14 or video or AI, it's a compelling story.
1:50:18 Like it's nothing changes. Nothing
1:50:22 changes except everything changes.
1:50:25 So the existing way of doing things in
1:50:29 every job imaginable is about to be
1:50:31 transformed.
1:50:34 All right. Uh we don't need James
1:50:36 Cameron anymore. Exactly, Robbie.
1:50:38 Exactly. Kyle, did you figure out the
1:50:41 private setting for lovable? Um I
1:50:43 didn't. Let me see if they got back to
1:50:45 me. I did put a request for someone to
1:50:49 [ __ ] help me on the lovable thing
1:50:52 questions.
1:50:55 Uhuh.
1:50:57 [Music]
1:51:00 Oh, they're never going to answer
1:51:02 everything because now that they're
1:51:03 doing this contest.
1:51:07 No, they didn't get back to me. Let me
1:51:09 go look.
1:51:28 project cardboard. Let's see if they
1:51:30 fixed the bug.
1:51:33 So, I'm going here. I'm going to
1:51:35 settings. It's got workspace. I'm going
1:51:38 to change it to public. I just changed
1:51:40 it. says visibility updated. Your
1:51:43 project is now public. Now I'm going to
1:51:45 switch out of this page
1:51:48 and come back to it. And it still says
1:51:50 workspace. I do not how I do not know
1:51:54 how to make this setting stick.
1:51:59 I don't
1:52:02 I've since quit and restarted my machine
1:52:06 and
1:52:08 Chrome.
1:52:13 The only thing I could think is I go
1:52:15 settings,
1:52:18 public,
1:52:22 publish,
1:52:25 update.
1:52:35 Project updated.
1:52:37 Try it now.
1:52:44 Brandon, try it now.
1:52:48 If it if it's still not there, like I I
1:52:50 don't I don't know.
1:52:55 Gaffin Biddy, some of us has have lost
1:52:57 interested in people. Nothing. I don't
1:53:00 know what to say. Brandon, I I um I put
1:53:03 a request into the lovable Discord and
1:53:06 they didn't get back to me. I'm gonna I
1:53:09 think I'm going to DM Tisha.
1:53:13 I'll I'll work on that this weekend.
1:53:14 I'll see if I can get that fixed this
1:53:15 weekend. Um
1:53:18 All right. Let's go back to Flo. Okay.
1:53:20 Now we have something.
1:53:22 What I was hoping to do was have a video
1:53:24 we could put up on the the internets.
1:53:36 [Music]
1:53:38 This is pretty good. Hey, Giants, I'm 6
1:53:41 in of pure doom. Kneel before me,
1:53:43 Redwoods, now hand me a ladder. This
1:53:45 lens insults my stature. Hey, Giants,
1:53:48 I'm 6 in of pure doom. Kneel before me,
1:53:51 Redwoods, now hand. That's pretty good.
1:53:53 All right, download. And then we'll
1:53:56 upres this to 1080p.
1:54:01 Okay. So now what I promised was
1:54:08 in irregulars.
1:54:12 I'm going to upload
1:54:21 share cool things. I'm going to upload
1:54:24 here is the
1:54:28 PDF.
1:54:30 of
1:54:33 the
1:54:36 um deep research
1:54:42 guide to VO3.
1:55:00 All right.
1:55:02 So, don't notify there. So, that's
1:55:05 there. And then in AI tool talk, which
1:55:08 is behind the mastermind firewall,
1:55:13 I'm going to share
1:55:15 the project. So, let me go to the
1:55:17 project
1:55:21 video scene builder.
1:55:27 So, can you share a project? Maybe you
1:55:29 can't. Hang on.
1:55:34 I can rename it. I can delete it.
1:55:38 I don't think I can share it.
1:55:43 H.
1:55:47 But what I'll put in what I'll put in
1:55:48 tool talk is I'll put that that thing
1:55:51 that we actually put some time into.
1:55:55 Oh, wait. You're not seeing any of this.
1:55:57 God damn that thing.
1:56:08 I find the GPTs I want, but I can't
1:56:10 invoke it on my chat.
1:56:13 I don't think you can make a GPT if
1:56:15 you're just cosmic. If you're just using
1:56:18 the mobile app, I don't think you can
1:56:20 create a GPT on the mobile app. I think
1:56:23 only on desktop.
1:56:29 No export option. No, there's nothing.
1:56:32 Let me Oh, uh, let me go to a new chat.
1:56:38 Um
1:56:41 in chat GPT
1:56:45 projects,
1:56:46 can you share a project
1:56:51 with the public?
1:56:58 You cannot currently directly share a
1:57:00 project. All right. So, I'll just go to
1:57:03 um to tool talk and I'll say here is the
1:57:07 VO3
1:57:10 prompting structured prompting guide.
1:57:14 Structured
1:57:17 prompting
1:57:20 prompting
1:57:22 guide. And I'm going to put um with
1:57:26 acting
1:57:29 instructions
1:57:30 and dialogue.
1:57:36 Um,
1:57:38 this
1:57:40 is good
1:57:42 to use as a knowledge
1:57:46 file for a
1:57:49 project in chat GPT
1:57:53 to generate
1:57:56 prompts
1:57:58 without
1:58:00 having to use your brain. Yay.
1:58:07 All right. And so I'll put the recipe
1:58:10 and filmmaker guide in there. This
1:58:12 thing's This thing's really cool.
1:58:13 Actually, the one thing that I would fix
1:58:16 about this is I would fix it so it
1:58:17 doesn't add quote marks to dialogue.
1:58:20 All right. And then we've got Let's go
1:58:23 get our file.
1:58:26 Did this download our our dude?
1:58:29 No.
1:58:31 Download. Okay, there we go. So, let's
1:58:34 go look at
1:58:42 Oh, he looks great. Okay, so this is the
1:58:45 1080p version
1:58:48 of our dude.
1:58:51 Hey, giants. I'm 6 in of pure doom.
1:58:54 Kneel before me, Redwoods. Now, hand me
1:58:56 a ladder. This lens insults my stature.
1:59:00 Nice. All right. Beautiful. Beautiful
1:59:04 peoples.
1:59:05 So let's go to
1:59:09 X.
1:59:10 X marks the spot.
1:59:13 Compose post.
1:59:16 Yes. Rather
1:59:22 the pilot
1:59:25 of the
1:59:27 Micro Foot
1:59:30 series.
1:59:35 Big foots
1:59:38 are overrated.
1:59:45 Micro foots
1:59:48 are the new
1:59:54 are all the rage.
2:00:02 #Bigfoot
2:00:10 Bigfoot AI. There we go. All right. And
2:00:13 so we're going to put up
2:00:19 [Music]
2:00:25 the new version of Whisk should be
2:00:26 available for everyone. It's fun. Okay.
2:00:30 Yeah, I think I saw that. Is it that's
2:00:32 at is it at uh labs labs.google.com?
2:00:39 There's flow.
2:00:58 Create
2:01:01 flow sparkify video whisk. Try it now.
2:01:05 New experimental tool that lets you use
2:01:06 images as prompts. Visual visual ideas
2:01:10 to tell your stories.
2:01:12 Enter the tool.
2:01:21 That's cool. They added a little guide
2:01:22 to the beginning of this. This is pretty
2:01:24 slick.
2:01:27 All right. So, you put in a picture
2:01:28 image, a place image, and whatever that
2:01:32 one is.
2:01:35 Style. Oh, cool. So, you can either
2:01:38 enter Oh, they've really improved the
2:01:39 the interface here. This is this is
2:01:42 slick.
2:01:44 Yeah. So, if you go to
2:01:45 labs.googlefxtoolswisk
2:01:46 Googlefxtoolswhisk
2:01:51 here. I'll put it in the I'll put it in
2:01:52 the the YouTube thing
2:02:00 authentically. Joe Ultra Plan has V2
2:02:02 credits in Whisk as well. Oh, cool.
2:02:04 Nice.
2:02:06 All right. Groovy G. Groovy Groovy
2:02:10 Groovy. I like it. All right. Did that
2:02:13 upload
2:02:16 error. No. Why would it upload? Why,
2:02:20 Elon? Why? Why?
2:02:23 Why can I not upload a [ __ ] file to
2:02:26 your stupid [ __ ] service?
2:02:31 261
2:02:33 error.
2:02:36 Well, let's try posting it. Post
2:02:41 2 24 40 50 60.
2:02:46 Come on, you can do it.
2:02:57 Success.
2:02:59 All right, everyone. Go to my thing. Go
2:03:01 find Bigfoot. Let's start the microfut
2:03:04 trend.
2:03:05 [Laughter]
2:03:14 All right, everybody.
2:03:16 Everybody. Everybody,
2:03:20 I'm going to get out of here.
2:03:23 Have a fantastic weekend. Um, here's the
2:03:26 thing.
2:03:28 So, for homework this weekend, um,
2:03:32 if you were here for most of the night,
2:03:34 like what I experimented with is going
2:03:37 from
2:03:39 just prompting informally to creating a
2:03:42 more structured tool to help you prompt.
2:03:45 Like, I would just play around with
2:03:46 that. Just play around with the idea of
2:03:49 any place where you're sort of
2:03:52 halfassing your prompt, get chat GPT to
2:03:56 help you do that in a in a more
2:03:58 sophisticated way, like whatever that
2:04:00 means.
2:04:02 Um, discover the power of unplugging
2:04:05 your brain and letting it do the hard
2:04:07 work and just surfing at this higher
2:04:09 level. Um, I I I think there's some real
2:04:13 some very deep deep deep
2:04:16 power there. And I think in the future
2:04:18 as these tools get more sophisticated,
2:04:21 they're going to naturally do a lot of
2:04:23 the stuff for us. We we just won't even
2:04:25 know what's happening on the back end
2:04:26 and it'll be doing all this anyway. But
2:04:28 if you want to if you want some of the
2:04:30 advantages of that now, you have to you
2:04:33 have to dig a bit deeper. So that might
2:04:35 be something fun to play with this
2:04:36 weekend. All right. Um happy Father's
2:04:39 Day. Thank you very much. Um hope you
2:04:41 have a fantastic weekend. I will be back
2:04:43 Monday. Um, next week we've got uh the
2:04:49 AI salon um meet and greet. So, if you
2:04:54 want to meet people in this community,
2:04:56 come to those meetings. Getting to know
2:04:59 people in this community is incredibly
2:05:01 important. The meeting we had today uh
2:05:03 in the mastermind, the mastermind um
2:05:05 live session that we have on Fridays now
2:05:08 um was a conversation of a different
2:05:10 nature. It was it was a much more
2:05:12 specific, much more tactical people
2:05:15 really helping one another. Like I
2:05:16 really like what's starting to already
2:05:18 emerge there. We just started it. So
2:05:20 super excited about that. Um all right,
2:05:24 have a fantastic weekend everyone. Happy
2:05:26 Father's Day to everybody and I will see
2:05:29 you next week.